mcmcensemble 3.2.0

Bug fixes

a bug resulting in weak correlation across walkers has been resolved (#9, @scheidan). Runs from this version will therefore produce results different than with previous versions, even for a given seed.

Two changes ensure chains don’t get stuck in a loop:

an early parameter check ensures that sufficient walkers are provided (via the n.walkers argument) to ensure ergodicity (#11, @scheidan)
noise has been added to each step of the differential evolution algorithm to ensure we don’t end up walking on a grid (#12, @scheidan)

there is now a clearer error message when trying to use a single walker since ensemble sampling is designed to work with multiple walkers (#6 by @Bisaloo, based on a report from @adamkucharski).

the arguments lower.inits and upper.inits are deprecated in favour of inits which leave more flexibility to the user. Please read the detailed blog post for more background about this change and how to migrate.
inits can now be a data.frame or a matrix
d.e.mcmc() and s.m.mcmc() are not exported any more. Please use the wrapper MCMCEnsemble() instead.
there is a new vignette (vignette("diagnostic-pkgs", package = "mcmcensemble")) presenting two different options (coda and bayesplot) to plot and evaluate the MCMC chains produced by mcmcensemble.

The chains now run fine even in the case where there is only one iteration (i.e., max.iter %/% n.walkers == 1)
The error message when the coda package is absent and coda = TRUE now correctly prompt the user to use coda = FALSE if they do not wish to install coda.

it is now possible to use a named vector as first argument of the function passed in f. This is useful if you do something like:

p.log.named <- function(x) {
  B <- 0.03
  return(-x["a"]^2/200 - 1/2*(x["b"]+B*x["a"]^2-100*B)^2)
}

mcmcensemble now explicitly depends on R >= 3.5.0. This was already implicitly the case since 2.1 because of the dependency on the progressr package.
the ensemble sampling algorithm used by MCMCEnsemble() is now recorded in an additional attribute (accessible via attr(res, "ensemble.sampler")).

there is now an additional argument check ensuring that lower.inits and upper.inits have the same names

the ensemble sampling can now be parallelised with the future framework. Check the README for more information

very large log.p differences between chains do not cause them to be stuck any more
addition of a new vignette listing frequently asked questions (with their answer)

The argument names and order in d.e.mcmc() and s.m.mcmc() now match those of MCMCEnsemble()